Axis 6: Digital Sciences & Artificial Intelligence

Marie-Dominique Devignes

  • Head Researcher, CNRS
  • Team Leader, CAPSID@ INRIA LORIA

Malika Smail

  • Asst. Prof. of Computer Sciences, Univ. of Lorraine
Digital sciences, Artificial Intelligence (AI), modeling of biological systems, and simulation constitute a cornerstone of the CARTAGE-PROFILES FHU project.
The amount and heterogeneity of patients’ data (multi-omic, imaging, clinical phenotypes, electronic health records) can be used to improve care and better understand pathophysiological process underlying patients’ diseases. AI methods will be developed including the use of biomedical 16 ontologies, on-line machine learning, text mining and temporal pattern mining to improve patient classification into relevant cardiovascular aging trajectories and for “dreaded event” prediction.
The main objective of the Digital Science axis is to develop and implement relevant mathematical and computational frameworks to set up AI-based strategies to better understand the cardiovascular aging process and improve personalized healthcare according to CARTAGE PROFILES scientific objectives. The long-term perspective is to build and experiment with “Learning Health Systems” that integrate on-line machine learning and knowledge discovery methods into the clinical life cycle thereby transforming collected patient data into improved healthcare settings.






Main Hypotheses

A. Various types of cardio-vascular aging profiles are embedded within the heterogeneous longitudinal clinical and biomedical data that can be collected in healthcare studies; data science and AI methods are needed to extract and characterize such profiles in the perspective of personalized medicine.

B. Modelling of pathophysiological processes with complex networks and various mathematical logical and probabilistic frameworks can contribute to simulate the onset and course of various diseases and understand the relative influence of a wide range of biological, clinical or drugrelated parameters in the aging process.

The originality and added value of the CARTAGE-PROFILES Digital Sciences axis rely in its long-standing original pluri-and transdisciplinary approaches, combining methods from basic science to targeted cohorts on CVRA and stroke. Importantly, the CARTAGE-PROFILES initiative brings together physicians involved in clinical management and clinical research and data scientists with strong expertise in AI and modelling of complex biological phenomena from the computer science laboratory LORIA and the mathematics laboratory IECL. On numerous occasions, senior and junior scientists from each concerned team have already met, discussed and worked together to apply new computer science and mathematical approaches to clinical and biological datasets, to interpret results and design new experiments.
The Digital Science axis is supported by the institutional environment of the concerned teams: UL, CNRS and Inria for LORIA and IECL; CHRU, UL and Inserm for CIC-P. Moreover, two scientists (one CNRS and one UL) benefit from an interface contract with CHRU Nancy. Current funding includes different systems of excellence like Lorraine Université d’excellence (LUE) with the Impact Project GEENAGE (two postdocs) and the ANR PIA RHU Fight-HF. Hardware and equipment expenses are funded mainly by CPER IT2MP. The Digital Science axis will work in coordination with the Grand-Est regional plan supporting research in AI and health, and with national and international teams with which we are already collaborating, for example in the framework of F-CRIN (Investissement d’Avenir) and the emerging national Health Data Hub.
Selected prior publications of the teams involved in the Digital Science axis:
  • Healthcare trajectory mining (81).
  • Detection of adverse drug effect associations using pattern structure and bio-ontologies (82).
  • Sequential linear regression with online standardized data to predict adverse events (83).
  • Similarity measurements based on unsupervised extremely randomized trees (84).
  • Exploitation of electronic health records for predicting drug-response variability (85).
  • Modeling of telomere length distribution (61).
  • Biomarker identification and interpretation in phenotypic subgroups at-risk for cardio-vascular disease (86).
  • Longitudinal patient positive profiles in the anti-phospholipid syndrome (87).
References
61. Toupance S, Villemonais D, Germain D, et al. Sci Rep. 2019;9:685.81. Jay N, Nuemi G, Gadreau M, Quantin C. BMC Med Inform Decis Making. 2013;13:130. 82. Personeni G, Bresso E, Devignes MD, et al. J Biomed Semantics. 2017;8:29. 83. Duarte, K., Monnez, J.M. and Albuisson, PloS One, 2018;13(1): e0191186 84. Dalleau K, Couceiro M, Smaïl-Tabbone M (22nd Pacific-Asia Conference on Knowledge Discovery and Data Mining), May 2018, Melbourne, Australia. 85. Coulet A, Shah NH, Wack M, et al. Scientific Reports 2018;8:15558. 86. Ferreira JP, Pizard A, Machu JLet al. Clin Res Cardiol. 2019;10.1007/s00392-019-01480-4. 87. Devignes J, Smaïl-Tabbone M, Hervé A, et al. Int J Lab Hematol. 2019;41:726-30